Restful sleep is essential for health, yet many children with Attention Deficit Hyperactivity Disorder (ADHD) experience disturbances such as delayed sleep onset, shorter total sleep time, frequent awakenings, and daytime fatigue. Accurate detection of these issues is important for clinical care, but existing tools have limitations: polysomnography is costly and complex, while wrist devices often miss subtle movement or physiological changes. This study introduces a deep learning approach using data from RestEaze, a leg-worn multimodal wearable that records photoplethysmography (PPG), motion from accelerometer and gyroscope, and temperature signals. Overnight recordings were collected from 14 children referred for ADHD evaluation. A Support Vector Machine (SVM) using handcrafted features was implemented to establish a traditional baseline. Two convolutional neural network (CNN-BiLSTM) models were then developed, employing early and late-fusion of raw multimodal inputs to classify sleep and wake states in short windows. The late-fusion model achieved an area under the ROC curve of 90.94% in five-fold cross-validation. Derived metrics included total sleep time, wake after sleep onset, sleep onset latency, and awakenings. A temporal label-smoothing method further improved consistency. These findings demonstrate the feasibility of leg-based multimodal sensing and deep learning for noninvasive sleep monitoring in pediatric neurodevelopmental populations.

Abstract Introduction Periodic limb movements during sleep (PLMS) are associated with repeated arousals and heart rate increases, suggesting sympathetic nervous system activation that could alter ventilation. Methods This retrospective, observational study evaluated breath-by-breath ventilation surrounding PLMS. NREM sleep epochs from polysomnograms with PLMS index (PLMI) ≥15 were analyzed. Validated signal processing algorithms analyzed EEG, leg EMG and nasal pressure data to identify PLMS, arousals, apneas, hypopneas and breath-by-breath ventilation. Ventilation (Vn) per breath was measured by integrating a square-root transformed nasal pressure signal over the breath’s duration, multiplying by respiratory rate and normalizing by 7-minute mean ventilation (baseVn). Vn=1.0 indicates ‘normal’ breathing; values above and below 1.0 indicate hyper- and hypoventilation. Each PLMS has an associated unique breath sequence: 6 breaths before (6B-1B), 1-2 breaths during (1D-2D), 8 breaths after (1A-8A) PLMS. Arousal was associated with PLMS when overlapping with 0.5 seconds before and after PLMS. Apnea or hypopnea were related to PLMS when overlapping with PLMS or breaths 1A-4A (PLMS-AH). PLMS-AH divided by hours NREM sleep (PLMS-AHI) ≥5 defined PLMS-related OSA (PLMS-rOSA). Results Seventy individuals (25 women) were 62.3±15.4 years old with BMI 32.5±8.4 kg/m2, median PLMI 73.3 and median AHI 14.8. 11 individuals had AHI< 5; 59 had AHI≥5. In total, there were 27,743 PLMS, 4,129 hypopneas and 2,679 apneas. 67.3% of apneas and hypopneas were PLMS-AH. 16.5% of PLMS were PLMS-AH; 31.9% for severe OSA. Median PLMS-AHI was 12.5. 78.6% of the cohort and 92.3% of OSA sufferers had PLMS-rOSA. Ventilation increased during PLMS and decreased following PLMS. For PLMS without arousal, Vn (medians) rose during PLMS to 1.07 (1D) and 1.05 (2D), while Vn fell after PLMS to a 0.91 nadir (4A). 1D-2D were significantly greater, while 2A-7A were significantly less than baseVn. For PLMS with arousal, Vn (medians) increased during PLMS to 1.35 (1D) and 1.40 (2D), and Vn fell to a 0.80 nadir (4A). 1D-2D and 1A were significantly greater, while 3A-8A were significantly less than baseVn. 39.6% of 4A breaths had Vn< 0.7. Conclusion Ventilation increased during PLMS and decreased following PLMS with larger magnitudes for PLMS with arousal. PLMS-rOSA was highly prevalent. Support (if any)

Cortical arousals are brief brain activations that disrupt sleep continuity and contribute to cardiovascular, cognitive, and behavioral impairments. Although polysomnography is the gold standard for arousal detection, its cost and complexity limit use in long-term or home-based monitoring. This study presents a noninvasive, machine learning–based framework for detecting cortical arousals using the RestEaze™ system, a leg-worn wearable that records multimodal physiological signals including accelerometry, gyroscope, photoplethysmography (PPG), and temperature. Across multiple methods tested, including logistic regression, XGBoost, and Random Forest classifiers, we found that features related to movement intensity were the most effective in identifying cortical arousals, while heart rate variability had a comparatively lower impact. The framework was evaluated in 14 children with attention-deficit/hyperactivity disorder (ADHD) undergoing assessment for restless leg syndrome–related sleep disruption. The Random Forest model achieved the best overall performance, with a ROC-AUC of 0.94 and an AUPRC of 0.55, substantially higher than the baseline prevalence of arousals ( 0.07). For the arousal class specifically, it reached a precision of 0.57, recall of 0.78, and F1-score of 0.65. These findings support the feasibility of wearable-based machine learning for real-world arousal detection, demonstrated here in a pediatric ADHD cohort with sleep-related behavioral concerns.

OBJECTIVES:Discrepancies between sleep diaries and sensor-based sleep parameters are widely recognized. This study examined the effect of showing sensor-based sleep parameters while completing a daily diary. The provision of sensor-based data was expected to reduce variance but not change the mean of self-reported sleep parameters, which would in turn align better with sensor-based data compared to a control diary. METHOD:In a crossover study, 24 volunteers completed week-long periods of control diary (digital sleep diary without sensor-based data feedback) or integrated diary (diary with device feedback), washout, and then the other diary condition. RESULTS:The integrated diary reduced self-reported total sleep time (TST) by <10 minutes and reduced variance in TST. The integrated diary did not impact mean sleep onset latency (SOL) and, unexpectedly, the variance in SOL increased. The integrated diary improved both bias and limits of agreement for SOL and TST. CONCLUSIONS:Integration of wearable, sensor-based device data in a sleep diary has little impact on means, mixed evidence for less variance, and better agreement with sensor-based data than a traditional diary. How the diary impacts reporting and sensor-based sleep measurements should be explored.

The cardiovascular system interacts continuously with the respiratory system to maintain the vital balance of oxygen and carbon dioxide in our body. The interplay between the sympathetic and parasympathetic branches of the autonomic nervous system regulates the aforesaid involuntary functions. This study analyzes the dynamics of the cardio-respiratory (CR) interactions using RR Intervals (RRI), Systolic Blood Pressure (SBP), and Respiration signals after first-order differencing to make them stationary. It investigates their variation with cognitive load induced by a virtual reality (VR) based Go-NoGo shooting task with low and high levels of task difficulty. We use Pearson's correlation-based linear and mutual information-based nonlinear measures of association to indicate the reduction in RRI-SBP and RRI-Respiration interactions with cognitive load. However, no linear correlation difference was observed in SBP-Respiration interactions with cognitive load, but their mutual information increased. A couple of open-loop autoregressive models with exogenous input (ARX) are estimated using RRI and SBP, and one closed-loop ARX model is estimated using RRI, SBP, and Respiration. The impulse responses (IRs) are derived for each input-output pair, and a reduction in the positive and negative peak amplitude of all the IRs is observed with cognitive load. Some novel parameters are derived by representing the IR as a double exponential curve with cosine modulation and show significant differences with cognitive load compared to other measures, especially for the IR between SBP and Respiration.

Anticipatory cardiac deceleration is the lengthening of heart period before an expected event. It appears to reflect preparation that supports rapid action. The current study sought to bolster anticipatory deceleration as a practical and unique estimator of performance efficiency. To this end, we examined relationships between deceleration and virtual reality performance under low and high time pressure. Importantly, we investigated whether deceleration separately estimates performance beyond basal heart period and basal high-frequency heart rate variability (other vagally influenced metrics related to cognition). Thirty participants completed an immersive virtual reality (VR) cognitive performance task across six longitudinal sessions. Anticipatory deceleration and basal heart period/heart period variability were quantified from electrocardiography collected during pre-task anticipatory countdowns and baseline periods, respectively. At the between-person level, we found that greater anticipatory declaration was related to superior accuracy and faster response times (RT). The relation between deceleration and accuracy was stronger under high relative to low time pressure, when good performance requires greater efficiency. Findings for heart period and heart period variability largely converge with the prior literature, but importantly, were statistically separate from deceleration effects on performance. Lastly, deceleration effects were detected using anticipatory periods that are more practical (shorter and more intermittent) than those typically employed. Taken together, findings suggest that anticipatory deceleration is a unique and practical correlate of cognitive-motor efficiency apart from heart period and heart period variability in virtual reality.

Abstract Introduction Obstructive Sleep Apnea (OSA) is characterized by reduction in airflow. Hypopnea is a smaller reduction in airflow compared to apnea but also accompanied by drop in oxygen saturation leading to sympathetic activation. Frequent OSA events are correlated with incidence of cardiovascular diseases. Measuring changes in airflow requires overnight polysomnography that is expensive. Electrocardiogram (ECG) is available through wearable devices at home and therefore, detecting OSA events using ECG can make OSA diagnostics more accessible. This paper studies the use of ECG morphology and an ensemble machine learning algorithm to detect OSA events. The algorithm designed is light weight to ensure that it can be implemented on an embedded processor. Methods The data from the Apnea, Bariatric surgery, and CPAP (ABC) study provided by the National Sleep Research Resource is used for the analysis. Twenty-six subjects diagnosed with severe OSA are considered with single night polysomnography before treatment. Since OSA events have a minimum 10s duration, 10s non-overlapping windows with OSA labels from a technician scoring is used. Features are extracted from the PQRST complex and majority voting across multiple algorithms is used to select the top 7 features (SDNN, RMSSD, P-P interval, P-duration, P-R interval, T-duration and T-P interval) with highest explanatory power. We train a random forest classifier using leave-one-out methodology where 25 subjects are used for training and the 26th subject is used for testing (all permutations are used). We use sensitivity, precision and F1 score for evaluation. Results The algorithm detects the precise occurrence (onset and offset) as well as the total number of events during the night. The precision, sensitivity and F1 scores are 70%, 96% and 80% respectively. The prediction leans towards higher occurrence of OSA events as the subjects suffer from severe OSA. We uncover 32% of false positives are events that are accompanied by significant SPO2 desaturation pointing to the inaccuracy of manual scoring. Assuming these false positives are correct predictions the precision increases to 79%. Conclusion The classification of OSA events using ECG features and machine learning demonstrates the feasibility of using wearables to measure OSA in a home setting. Support (if any)

Abstract Introduction Precision measurement of sleep metrics like sleep onset latency (SOL) and total sleep time (TST) has long been a challenge. Multiple studies have demonstrated significant discrepancies between sleep diaries and wearable device derived sleep metrics especially in patients with subjective-objective sleep discrepancy (SOSD; e.g., some patients with insomnia). Typically SOSD manifests as longer than expected self-reported SOL and shorter than expected TST. Without a method of reconciliation between sleep diaries and wearable devices, the current practice is to rely on sleep diary data which is suboptimal in the context of SOSD, and could contribute to blunted or nonsignificant effects in clinical trials. The current study investigated the effects of providing wearable device data feedback contemporaneously with the completion of the sleep diary, on a daily basis across 288 nights. We expected that by providing wearable device data feedback, participants with SOSD could be identified and potentially self-correct their SOSD. Methods A 3 phase, randomized, crossover design study was used in which 24 undergraduate college students without a diagnosed sleep disorder completed week-long periods of control condition (digital sleep diary without wearable device data feedback), washout, and then test phase (diary with device feedback). Participants were randomized to start with control or test conditions. Results Within and between subjects analyses were performed to understand the effect of the wearable device data feedback on sleep diary responses. Two participants (8.3%) with SOSD were identified and had marked differences in their estimated SOL (~47 min) that were corrected when provided with their wearable device data. Importantly no subjects copied their wearable device data into their sleep diary. Between subjects analyses revealed no differences in the average or variability of sleep metrics. Conclusion These preliminary results suggest that presenting wearable device data during sleep diary completion may impact self-reporting by individuals with SOSD but not by those without SOSD. In the setting of a clinical trial, participants who have significant SOSD may be contributing to non-disease/non-treatment related variability that may be ameliorated by providing wearable device data during sleep diary completion. Support (if any)

With the increased utilization, the small embedded and IoT devices have become an attractive target for sophisticated attacks that can exploit the devices security critical information and data in malevolent activities. Secure boot and Remote Attestation (RA) techniques verifies the integrity of the devices software state at boot-time and runtime. Correct implementation and formal verification of these security primitives provide strong security guarantees and enhance user confidence. The formal verification of these security primitives is considered challenging, as it involves complex hardware software interactions, semantics gaps and requires bit-precise reasoning. To address these challenges, this paper presents FVCARE an end to end system co-verification framework. It also defines the security properties for resilient small embedded systems. FVCARE divides the end to end system co verification problem into two modules: 1) verifying the (bit precise) initial system settings, registers, and access control policies by hardware verification techniques, and 2) verifying the system specification, security properties, and functional correctness using source-level software abstraction of the hardware. The evaluation of proposed techniques on SRACARE based systems demonstrates its efficacy in security co verification.

This work has been submitted to the IEEE Transactions on Affective Computing for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible. Overburdening an individual’s limited cognitive resources, especially when engaged in critical operations, may result in disastrous mishaps. Regular assessments of individuals’ physiological states and associated performance become vital to improving their mission readiness in such scenarios. As a key step towards a field-ready system, this treatise discusses the experimental findings pertinent to affective physiological state modulation and predictive modeling of marksmanship during a Go/NoGo shooting task in an immersive virtual reality environment. The shooting exercise requires the participants to hit the enemy targets and spare the friendly targets. The shooting difficulty levels (SDLs) are introduced by modulating the subject-specific target exposure time. The physiological signals used for analysis comprise electrocardiogram (ECG), 64-channel electroencephalogram (EEG), and standard shooting performance scores from 31 subjects. Experimental results with ECG features encompass involuntary physiologic process regulation and the interplay between the autonomic nervous system (ANS) components varying with SDL. Similarly, EEG features highlight the variations in brain region activations with SDLs. Predictive modeling of shooting performance (enemy hit, friendly spare, overall score) and behavioral response (mean enemy reaction time) from physiological (ECG and EEG) features evince the potency of physiological sensing for marksmanship estimation in operational contexts. Moreover, interpretable Shapley value analysis of the predictive models comprehend the (positive/negative) marginal impact of the underlying physiological features on marksmanship. This multimodal physiological sensing framework may assess the alterations in psychophysiological affective states and cognitive effects for performance analysis in operational contexts.

Objective.Most arrhythmias due to cardiovascular diseases alter the heart's electrical activity, resulting in morphological alterations in electrocardiogram (ECG) recordings. ECG acquisition is a low-cost, non-invasive process and is commonly used for continuous monitoring as a diagnostic tool for cardiac abnormality identification. Our objective is to diagnose twenty-nine cardiac abnormalities and sinus rhythm using varied lead ECG signals.Approach.This work proposes a deep residual inception network with channel attention mechanism (RINCA) for twenty-nine cardiac arrhythmia classification along with normal ECG from multi-label ECG signal with different lead combinations. TheRINCAarchitecture employing the inception-based convolutional neural network backbone uses residual skip connections with the channel attention mechanism. The inception model facilitates efficient computation and prevents overfitting while exploring deeper networks through dimensionality reduction and stacked 1-dimensional convolutions. The residual skip connections alleviate the vanishing gradient problem. The attention modules selectively leverage the temporally significant segments in a sequence and predominant channels for multi-lead ECG signals, contributing to the decision-making.Main results.Exhaustive experimental evaluation on the large-scale 'PhysioNet/Computing in Cardiology Challenge (2021)' dataset demonstratesRINCA's efficacy. On the hidden test data set,RINCAachieves the challenge metric score of 0.55, 0.51, 0.53, 0.51, and 0.53 (ranked 2nd, 5th, 4th, 5th and 4th) for the twelve-lead, six-lead, four-lead, three-lead, and two-lead combination cases, respectively.Significance.The proposedRINCAmodel is more robust against varied sampling frequency, recording time, and data with heterogeneous demographics than the existing art. The explainability analysis showsRINCA's potential in clinical interpretations.

Inadvertent use of medication that has been tampered with can cause serious harm. Monitoring how and when medication was last used or touched is important for mitigating risks. In this paper, we present a new radar-based monitoring system that can detect eight different types of tampering methods with three types of medication containers. Our system works by using a FMCW and CW Doppler radar to capture motion speed, direction, and range, which we use for classifying activities. For monitoring activities at home, our system can be set up underneath a kitchen cabinet to monitor medication left out on the countertop. As our system uses radar, we can preserve privacy of individuals as the signatures from the radar are specific to the locations of the antennas and not necessarily associated with an individual. For classifying activities we created a processing pipeline that extracts a set of features from the raw multivariate time series signals from the radar. We then used three types of data augmentation techniques including jittering, scaling, and magnitude warping, to increase our data sets and increase our classification model accuracy. In addition, we evaluated our system using 5-fold cross validation and with different types of augmentation data sets. Our system can achieve 99% accuracy using a logistic regression classifier with multiple people.

The hypothesis that the central nervous system (CNS) makes use of synergies or movement primitives in achieving simple to complex movements has inspired the investigation of different types of synergies. Kinematic and muscle synergies have been extensively studied in the literature, but only a few studies have compared and combined both types of synergies during the control and coordination of the human hand. In this paper, synergies were extracted first independently (called kinematic and muscle synergies) and then combined through data fusion (called musculoskeletal synergies) from 26 activities of daily living in 22 individuals using principal component analysis (PCA) and independent component analysis (ICA). By a weighted linear combination of musculoskeletal synergies, the recorded kinematics and the recorded muscle activities were reconstructed. The performances of musculoskeletal synergies in reconstructing the movements were compared to the synergies reported previously in the literature by us and others. The results indicate that the musculoskeletal synergies performed better than the synergies extracted without fusion. We attribute this improvement in performance to the musculoskeletal synergies that were generated on the basis of the cross-information between muscle and kinematic activities. Moreover, the synergies extracted using ICA performed better than the synergies extracted using PCA. These musculoskeletal synergies can possibly improve the capabilities of the current methodologies used to control high dimensional prosthetics and exoskeletons.

Leg movements during sleep occur in patients with sleep pathology and healthy individuals. Some (but not all) leg movements during sleep are related to cortical arousals which occur without conscious awareness but have a significant effect of sleep fragmentation. Detecting leg movements during sleep that are associated with cortical arousals can provide unique insight into the nature and quality of sleep. In this study, a novel leg movement monitor that uses a unique capacitive displacement sensor and 6-axis inertial measurement unit, is used in conjunction with polysomnography to understand the relationship between leg movement and electroencephalogram (EEG) defined cortical arousals. In an approach that we call neuro-extremity analysis, directed connectivity metrics are used to interrogate causal linkages between EEG and leg movements measured by the leg movement sensors. The capacitive displacement measures were more closely related to EEG-defined cortical arousals than inertial measurements. Second, the neuro-extremity analysis reveals a temporally evolving connectivity pattern that is consistent with a model of cortical arousals in which brainstem dysfunction leads to near-instantaneous leg movements and a delayed, filtered signal to the cortex leading to the cortical arousal during sleep.

Substance use disorder (SUD) is a dangerous epidemic that develops out of recurrent use of alcohol and/or drugs and has the capability to severely damage one's brain and behaviour. Stress is an established risk factor in SUD's development of addiction and in reinstating drug seeking. Despite this expanding epidemic and the potential for its grave consequences, there are limited options available for management and treatment, as well as pharmacotherapies and psychosocial treatments. To this end, there is a need for new and improved devices dedicated to the detection, management, and treatment of SUD. In this paper, the negative effects of SUD-related stress were discussed, and based on that, a few significant biomarkers were selected from a set of eight features collected by a chest-worn device, RespiBAN Professional, on fifteen individuals. We used three machine learning classifiers on these optimal biomarkers to detect stress. Based on the accuracies, the best biomarkers to detect stress and those considered as features for classification were determined to be electrodermal activity (EDA), body temperature, and a chest-worn accelerometer. Additionally, the differences between mental stress and physical stress, as well as different administrations of meditation during the study, were identified and analysed. Challenges, implications, and applications were also discussed. In the near future, we aim to replicate the proposed methods in individuals with SUD.

Stress is an established risk factor in the development of addiction and in reinstating drug seeking. Substance use disorder (SUD) is a dangerous epidemic that affects the brain and behavior. Despite this growing epidemic and its subsequent consequences, there are limited management and treatment options, pharmacotherapies and psychosocial treatments available. To this end, there is a need for new and improved personalized devices and treatments for the detection and management of SUD. Based on documented negative effects of stress in SUD, in this paper, our objective was to select a few significant physiological features from a set of 8 features collected by a chest-worn RespiBAN Professional in 15 individuals. We used three machine learning classifiers on these optimal physiological features to detect stress. Our results indicate that best accuracies were achieved when electrodermal activity (EDA), body temperature and chest-worn accelerometer were considered as features for the classification. Challenges, implications and applications were discussed. In the near future, the proposed methods will be replicated in individuals with SUD.

Physiological sensing of virtual reality (VR)-induced stressors are increasingly utilized to improve human training and assess the impact of gaming difficulty-induced stress on a person's health and well-being. However, the prior art sparsely explores the multi-level cardiovascular dynamics for psychophysiological demands in a VR environment. This treatise discusses the experimental findings and physiological interpretations of various heart rate variability (HRV) metrics extracted from 31 participants during a Go/No-Go VR-based shooting task across multiple timeframes. The VR-shooting exercise consists of firing at the enemy targets while sparing the friendly ones for different shooting difficulty levels: low-difficulty and high-difficulty with in-between baselines. Ex-perimental results demonstrate consistent shooting difficulty-induced stress patterns at multi-granular levels in response to the heterogeneous inputs (exogenous and endogenous factors). The physiological interpretations highlight the intricate inter-play between cardio-physiological components: sympathetic and parasympathetic response across multiple timescales (sessions and blocks) and shooting difficulty levels.

Several sleep disorders are characterized by periodic leg movements during sleep including Restless Leg Syndrome, and can indicate disrupted sleep in otherwise healthy individuals. Current technologies to measure periodic leg movements during sleep are limited. Polysomnography and some home sleep tests use surface electromyography to measure electrical activity from the anterior tibilias muscle. Actigraphy uses three-axis accelerometers to measure movement of the ankle. Electromyography misses periodic leg movements that involve other leg muscles and is obtrusive because of the wires needed to carry the signal. Actigraphy based devices require large amplitude movements of the ankle to detect leg movements (missing the significant number of more subtle leg movements) and can be worn in multiple configurations precluding precision measurement. These limitations have contributed to their lack of adoption as a standard of care for several sleep disorders. In this study, we develop the RestEaze sleep assessment tool as an ankle-worn wearable device that combines capacitive sensors and a six-axis inertial measurement unit to precisely measure periodic leg movements during sleep. This unique combination of sensors and the form-factor of the device addresses current limitations of periodic leg movements during sleep measurement techniques. Pilot data collected shows high correlation with polysomnography across a heterogeneous participant sample and high usability ratings. RestEaze shows promise in providing ecologically valid, longitudinal measures of leg movements that will be useful for clinicians, researchers, and patients to better understand sleep.

This paper explores power spectrum-based features extracted from the 64-channel electroencephalogram (EEG) signals to analyze brain activity alterations during a virtual reality (VR)-based stressful shooting task, with low and high difficulty levels, from an initial resting baseline. This paper also investigates the variations in EEG across several experimental sessions performed over multiple days. Results indicate that patterns of changes in different power bands of the EEG are consistent with high mental stress levels during the shooting task compared to baseline. Although there is one inconsistency, overall, the brain patterns indicate higher stress levels during high difficulty tasks than low difficulty tasks and in the first session compared to the last session.